Silhouette 2026 lands with fewer steps between “I need a clean matte” and “shot approved”, without taking the steering wheel away from artists. The release updates multiple ML-driven matting tools that can detect, segment, propagate, and refine, while still keeping the workflow artist-controlled.

The additions cluster around three pain points that never die in production: edge stability over time, tracking that survives real-world footage, and pipeline plumbing that does not explode when you scale beyond a single workstation. Silhouette 2026 touches all three, and it does it in a very node-graph way: new models, new nodes, and more data moving through the graph instead of getting stranded in side files.

AI matting that thinks in objects, not in one mask at a time

The AI toolset gets updates across five ML nodes, with a focus on scalable multi-object workflows. Face ML gets a new model plus new Nose and Neck matte options, a full-face matte option, and deeper integration with ML masking and compositing workflows. Depth Map ML gets a new model aimed at more detail and improved temporal consistency.

Mask ML adds multi-object workflow features, including object list management, foreground and background refinement tools, and per-object Cryptomatte data. Matte Assist ML adds intermediate frame caching, improved propagation performance across long shots, and multi-object Cryptomatte output. Matte Refine ML adds a new Video Matting model for improved temporal consistency along edges, and it preserves Cryptomatte data after refinement.

Be wary: running many ML masking nodes can load multiple custom models onto the GPU and fill VRAM quickly, so the multi-object approach inside a single Mask ML node matters for both speed and sanity.

If you live in the land of shallow depth of field, it is simple: keep your IDs, then soften your edges with the model that holds up over time. The Video Matting model option in Matte Refine ML is a temporal-consistency favourite in the training material.

Head Track ML and the 3D Scene node get real

Silhouette’s 3D Scene node now pushes further into 3D-aware shot work. It is powered by SynthEyes technology, and Silhouette 2026 adds a new workflow centrepiece: Head Track ML.

Head Track ML uses ML to automatically build and track a full 3D head mesh, preserving facial motion and expressions for face-aware effects. The ML model loads onto the user’s graphics card, detects the head, and then Silhouette creates and tracks an aligned and scaled 3D head mesh.

From there, the new Unwrap and Rewrap nodes connect 3D to 2D UV space for roto, paint, and beauty work, and a Render Depth node can generate depth maps from Head Track ML meshes for beauty work or downstream VFX pipelines.

The training walk-through frames the workflow as a practical alternative to the classic “stabilize a patch, paint, reapply motion, repeat until retirement” approach. The point is to unwrap the tracked head into a stable 2D space, do the detail work, then rewrap it back onto the moving performance. It even comes with a candy-wrapper analogy, because every 3D workflow deserves at least one metaphor before it hits the studio floor.

There’s limits: For example, Head Track ML does not detect relative distance between multiple heads, and suggests using an “All render” option in the Render Depth node if a depth map for all heads is still needed.

On the camera side, Refine Solve is designed to let artists improve a camera solve without reprocessing the entire solve. It supports fixing bad frames and smoothing problem areas by deleting erratic or unnecessary 3D features, aiming for more stable solves and faster cleanups.

Tracking upgrades that target the annoying bits of reality

Silhouette 2026 gives love to two tracking tasks that tend to chew time: point tracks that fall apart when anything gets in the way, and repetitive object tracks for faces and license plates.

Point Track ML updates the long-running point tracker workflow with ML assistance aimed at motion blur, occlusions, objects going off-screen, and changing reference patterns. It includes three accuracy settings and Viewer Stabilisation to validate and lock the track. This is not a guaranteed fix, but it increases the chance of success in those failure cases that always show up right before delivery.

Object Tracker detects and creates layers for object classes, including faces and license plates, to reduce setup time, and tracker types support occlusion matte input. In the tutorial flow, it can automatically generate new layers and shapes when new faces enter a shot during a forward track.

This feature that works best when you treat it like a fast first pass – the result may not be 100 per cent perfect, but perfect for fast turnaround work. You don’t get “final shapes”, but points where you can attach additional Tracking data, and modify what needs to be cleaned.

Paint tools that aim for fewer clicks and fewer regrets

Silhouette’s paint system gets new tools and workflow changes aimed at cleanup and beauty work. A new Healing Brush supports cleanup and retouching with expanded colour sampling from any input. Brush controls add blend modes and pressure-sensitive sizes. Blur, Clone, and Drag brushes get more natural and flexible behaviour. Clone workflow updates add continuous mode and auto-grade locking. Resize Paint introduces a resolution-independent workflow that allows working in a lower-resolution session, then rebuilding strokes accurately on higher-resolution media.

Resize Paint is a way to work quickly at low resolution and later rebuild exact stroke replicas at high resolution, which is basically a polite way of saying your workstation fans can stop auditioning for a jet engine while you block in cleanup.

Cryptomatte, data ports, and fewer matte files to babysit

Silhouette 2026 deepens Cryptomatte-centric workflows, and the important bit is that it treats mattes as data that should travel with the graph, not as files that should multiply in folders.

Pipeline changes include passing filtered Cryptomatte data as alpha through nodes via an Matte input, enabling objects and layers to act as matte sources. A new Cryptomatte Mixer node can combine multiple Cryptomatte streams into a single output. ML and AI tools support multi-object outputs with improved viewer-based selection tools to speed up scene isolation and matte management.

Beware the workflow trap: do not build one Mask ML and one Matte Assist ML node per object, because each node loads its own model to the GPU and VRAM fills fast. Instead, multiple objects can live inside a single Mask ML node, and Cryptomatte over the data port can move reference data to Matte Assist ML even when reference keyframes are scattered along the timeline.

For OFX plugin users, there is a limitation: Cryptomatte data cannot simply be routed to an output node and expected to appear directly in the host application, due to how OFX plugins work. The workflow is to export to the file system instead, using an Output Multi node for Cryptomatte-generating nodes.

Export details matter, too. Cryptomatte needs EXR for multi-channel properties and uses only lossless compression, and it warns against lossy formats such as DWAA, DWAB, or PXR24. That is a lot of words to say: keep your IDs clean, keep them in one sequence, and do not turn your matte export into a science project with ten folders named “final_final_3_forreal_v02”. Unless you enjoy that sort of thing. Some people collect stamps, after all.

Performance, stability, and the stuff you notice at 2 a.m.

Silhouette 2026 includes performance and playback improvements through ML node caching for several ML nodes, plus predictive playback via pre-fetching and batched render jobs. Playback controls add optimisation options, including adjustable job queue size, preload frames, and threading settings.

Daily workflow improvements include advanced node search with category, intent, and data-type filtering, a Favourites node, an improved naming system, a context-sensitive HUD, and a Safe Mode troubleshooting workflow to enhance stability in demanding environments.

This is the kind of feature that helps the person who inherits your comp at the worst possible time. If your node tree looks like it was assembled during an earthquake, you still have to deliver, so better search and safer startup modes are not flashy, but deeply production-friendly. Yes, you should still test new tools and innovations before use in production, especially when ML models and GPU caching sit in the critical path.

Pipeline automation and AI assistants, now with actual plumbing

The most pipeline-forward change is a built-in JSON-RPC server and an MCP server. The JSON-RPC server allows studios to drive Silhouette remotely, execute Python code, run script files, and control the application via JSON-RPC.

A Python Package Manager manages third-party Python packages inside a directory in Silhouette and stays isolated from the system Python. This is powerful, and it is also the kind of feature that deserves a boring rollout plan. Build guardrails, keep permissions sane, and do not let the intern’s chatbot become the new show supervisor. The intern already has enough power.

The MCP server allows direct calls from AI assistants, which can inspect projects, build node trees, edit shapes and keyframes, and render frames. “ChatGPT, explain what I did there!” is now a valid question that can be answered. Not truthfully or understandably, because … LLMs, but still.

What to watch to get productive fast

If you want a training rabbit hole, the Silhouette training and tutorial ecosystem lives in the video library and training hub on the BorisFX-site, spanning both quick tips and longer courses on all packages and levels of experience.

Pricing and availability

Silhouette is available as a standalone application and as a plugin, with subscription, upgrade, support, and permanent license options. Subscriptions start at $ 545 per year or $ 103 per month. Silhouette supports macOS, Windows, and Linux. Supported host applications include Adobe After Effects, Adobe Premiere, Foundry Nuke, Blackmagic Resolve and Fusion, and Autodesk Flame.

Customers with an active Silhouette subscription, Boris FX Suite subscription, or an active Silhouette upgrade and support plan receive the 2026.5 release directly as a update.


https://borisfx.com/products/silhouette/

The post Silhouette 2026 gets smarter roto and 3D heads first appeared on DIGITAL PRODUCTION and was written by Bela Beier.